I'm looking for an easy way (a command or series of commands, probably involving find) to find duplicate files in two directories, and replace the files in one directory with hardlinks of the files in the other directory.

Here's the situation: This is a file server which multiple people store audio files on, each user having their own folder. Sometimes multiple people have copies of the exact same audio files. Right now, these are duplicates. I'd like to make it so they're hardlinks, to save hard drive space.

  • 25
    One problem you may run into with hardlinks is if somebody decides to do something to one of their music files that you've hard-linked they could inadvertently be affecting other people's access to their music.
    – Steven D
    Commented Oct 13, 2010 at 2:48
  • 4
    another problem is that two different files containing "Some Really Great Tune", even if taken from the same source with the same encoder will very likely not be bit-for-bit identical.
    – msw
    Commented Oct 13, 2010 at 2:57
  • 4
    better sollution might be to have a public music folder...
    – Stefan
    Commented Oct 13, 2010 at 7:08
  • 4
    related: superuser.com/questions/140819/ways-to-deduplicate-files
    – David Cary
    Commented Mar 16, 2011 at 23:59
  • 3
    @tante: Using symlinks solves no problem. When a user "deletes" a file, the number of links to it gets decremented, when the count reaches zero, the files gets really deleted, that's all. So deletion is no problem with hardlinked files, the only problem is a user trying to edit the file (unprobable indeed) or to overwrite it (quite possible if logged in).
    – maaartinus
    Commented Mar 14, 2012 at 3:56

20 Answers 20


rdfind does exactly what you ask for (and in the order johny why lists). Makes it possible to delete duplicates, replace them with either soft or hard links. Combined with symlinks you can also make the symlink either absolute or relative. You can even pick checksum algorithm (sha256, md5, or sha1).

Since it is compiled it is faster than most scripted solutions: time on a 15 GiB folder with 2600 files on my Mac Mini from 2009 returns this

9.99s user 3.61s system 66% cpu 20.543 total

(using md5).

Available in most package handlers (e.g. MacPorts for Mac OS X).

Edit: I can add that rdfind is really easy to use/very pedagogic. Just use the -dryrun true flag and it will be very intuitive, not scary (which, IMO, tools that delete files usually are).

  • 18
    +1 I used rdfind and loved it. It has a -dryrun true option that will let you know what it would have done. Replacing duplicates with hard links is as simple as -makehardlinks true. It produced a nice log and it let me know how much space was freed up. Plus, according to the author's benchmark, rdfind is faster than duff and fslint. Commented Dec 29, 2013 at 20:49
  • 10
    I suspect the performance of this tool has more to do with the algorithm itself and less to do with whether it's a compiled tool or a script. For this kind of operation, disk is going to be the bottleneck nearly all of the time. As long as scripted tools make sure that they've an async I/O operation in progress while burning the CPU on checksums, they should perform about as well as a native binary.
    – cdhowie
    Commented May 31, 2018 at 21:19
  • 10
    Beware that some versions of rdfind are affected by a pretty inconvenient bug : source files are deleted when hardlinks can't be created. Commented Jan 2, 2020 at 17:01
  • 5
    @d-b I wonder what proportion of systems are less than 1.5 years up-to-date. I lost data, and other people will if they don't check their version. They deserve to be warned. Commented Mar 23, 2020 at 10:08
  • 2
    @d-b -- I have to agree with Skippy; I am running the latest Mint with all updates and still have rdfile 1.3.5, rdfile 1.4.0 is where the fix is at according to the change log. I lost data (annoying, but easily recovered from backups).
    – Nathanael
    Commented Jun 12, 2020 at 6:03

Use the fdupes tool:

fdupes -r /path/to/folder gives you a list of duplicates in the directory (-r makes it recursive). The output looks like this:



with filename1 and filename2 being identical and filename3, filename4 and filename5 also being identical.

  • 1
    Ubuntu Note: As of September 2013, it hasn't had a stable release (it is on 1.50-PR2-3), so the update doesn't appear in ubuntu yet. Commented Aug 28, 2013 at 14:19
  • 11
    I just tried installing fdupes_1.50-PR2-4 on both Ubuntu and Debian, neither has the -L flag. Luckily building from github.com/tobiasschulz/fdupes was super easy.
    – neu242
    Commented Aug 30, 2013 at 15:07
  • 3
    Try rdfind - like fdupes, but faster and available on OS X and Cygwin as well.
    – oligofren
    Commented Jan 3, 2015 at 13:43
  • 14
    fdupes seems to only find duplicates, not replace them with hardlinks, so not an answer to the question IMO.
    – Calimo
    Commented Nov 8, 2017 at 15:58
  • 7
    There's a similar tool called jdupes that's based on fdupes, but it can also replace the duplicate files with symlinks (-l), hardlinks (-L) or instruct btrfs to deduplicate the blocks on the filesystem level (-B, if you're using btrfs). Commented Aug 22, 2018 at 16:36

There is a perl script at http://cpansearch.perl.org/src/ANDK/Perl-Repository-APC-2.002/eg/trimtrees.pl which does exactly what you want:

Traverse all directories named on the command line, compute MD5 checksums and find files with identical MD5. IF they are equal, do a real comparison if they are really equal, replace the second of two files with a hard link to the first one.

  • Sounds perfect, thanks!! I'll try it and accept if it works as described!
    – Josh
    Commented Oct 12, 2010 at 20:09
  • 3
    This did exactly what I asked for. However I believe that ZFS with dedup will eventually be the way to do, since I did find that the files had slight differences so only a few could be hardlinked.
    – Josh
    Commented Dec 8, 2010 at 20:13
  • 15
    Upvoted this, but after researching some more, I kind of which I didn't. rdfind is available via the package managers for ALL major platforms (os x, linux, (cyg)win, solaris), and works at a blazing native speed. So do check out the answer below.
    – oligofren
    Commented Jan 3, 2015 at 13:42
  • @oligofren I was thinking the same, but then I hit [Errno 31] Too many links. This scrips seems to be the only thing that handles that.
    – phunehehe
    Commented Jun 26, 2015 at 6:59
  • 8
    Checksumming every single file, rather than only files where there exists at least one other with identical size, is unnecessarily inefficient (and unnecessarily prone to hash collisions). Commented Feb 1, 2016 at 16:56

I use hardlink from http://jak-linux.org/projects/hardlink/

  • 1
    Nice hint, I am using on a regular base code.google.com/p/hardlinkpy but this was not updated for a while...
    – meduz
    Commented Apr 11, 2012 at 19:09
  • 2
    This appears to be similar to the original hardlink on Fedora/RHEL/etc.
    – user12810
    Commented Jun 21, 2012 at 8:43
  • 3
    hardlink is now a native binary in many Linux package systems (since ~2014) and extremely fast. For 1,2M files (320GB), it just took 200 seconds (linking roughly 10% of the files). Commented Feb 5, 2017 at 19:13
  • FWIW, the above hardlink was created by Julian Andres Klode while the Fedora hardlink was created by Jakub Jelinek (source: pagure.io/hardlink - Fedora package name: hardlink) Commented Jan 4, 2019 at 17:52

This is one of the functions provided by "fslint" -- http://en.flossmanuals.net/FSlint/Introduction

Click the "Merge" button:


  • 4
    The -m will hardlink duplicates together, -d will delete all but one, and -t will dry run, printing what it would do
    – Azendale
    Commented Oct 29, 2012 at 5:57
  • 1
    On Ubuntu here is what to do: sudo apt-get install fslint /usr/share/fslint/fslint/findup -m /your/directory/tree (directory /usr/share/fslint/fslint/ is not in $PATH by default)
    – Jocelyn
    Commented Sep 8, 2013 at 15:38

Since your main target is to save disk space, there is another solution: de-duplication (and probably compression) on file system level. Compared with the hard-link solution, it does not have the problem of inadvertently affecting other linked files.

ZFS has dedup (block-level, not file-level) since pool version 23 and compression since long time ago. If you are using linux, you may try zfs-fuse, or if you use BSD, it is natively supported.

  • This is probably the way I'll go eventually, however, does BSD's ZFS implementation do dedup? I thought it did not.
    – Josh
    Commented Dec 8, 2010 at 20:14
  • 22
    ZFS dedup is the friend of nobody. Where ZFS recommends 1Gb ram per 1Tb usable disk space, you're friggin' nuts if you try to use dedup with less than 32Gb ram per 1Tb usable disk space. That means that for a 1Tb mirror, if you don't have 32 Gb ram, you are likely to encounter memory bomb conditions sooner or later that will halt the machine due to lack of ram. Been there, done that, still recovering from the PTSD.
    – killermist
    Commented Sep 22, 2014 at 18:51
  • 7
    To avoid the excessive RAM requirements with online deduplication (i.e., check on every write), btrfs uses batch or offline deduplication (run it whenever you consider it useful/necessary) btrfs.wiki.kernel.org/index.php/Deduplication Commented Feb 5, 2017 at 19:18
  • 8
    Update seven years later: I eventually did move to ZFS and tried deduplication -- I found that it's RAM requirements were indeed just far to high. Crafty use of ZFS snapshots provided the solution I ended up using. (Copy one user's music, snapshot and clone, copy the second user's music into the clone using rsync --inplace so only changed blocks are stored)
    – Josh
    Commented Sep 13, 2017 at 13:54
  • 1
    @endolith Nope! Which makes ZFS dedup completely useless for 99% of users. If you have enough RAM for online dedup, either your disks are tiny or your wallet is big enough that you should instead spend some engineer-hours implementing dedup in your application.
    – Navin
    Commented Jan 20, 2022 at 13:10

On modern Linux these days there's https://github.com/g2p/bedup which de-duplicates on a btrfs filesystem, but 1) without as much of the scan overhead, 2) files can diverge easily again afterwards.

apt show hardlink

Description: Hardlinks multiple copies of the same file Hardlink is a tool which detects multiple copies of the same file and replaces them with hardlinks.

I also used jdupes recently with success.

  • The only program mentioned here available for Gentoo without unmasking and with hardlink support, thanks! Commented Mar 9, 2015 at 13:48

jdupes has been mentioned in a comment but deserves its own answer, since it is probably available in most distributions and runs pretty fast (it just freed 2.7 GB of a 98 % full 158 GB partition (SSD drive) in about one minute) :

jdupes -rL /foo/bar
  • 1
    jdupes really should be the answer. It's fast, and configurable, and it shows progress as you go. You can live dangerously (-T -T) when initially scanning without modifying (-M) to see what you might be in for. Then choose the default size / partial hash / full hash / bit by bit comparison mode, skip the hashes and just go straight to bits (-K) without any added risk, or roll the dice and do the full hash only while skipping the bit comparison (-Q). About the only thing missing is an interactive mode where you could y/n each suggestion. Did I mention that it's fast?
    – Chris
    Commented Jun 12, 2020 at 0:36

To find duplicate files you can use duff.

Duff is a Unix command-line utility for quickly finding duplicates in a given set of files.

Simply run:

duff -r target-folder

To create hardlinks to those files automaticly, you will need to parse the output of duff with bash or some other scripting language.


Seems to me that checking the filename first could speed things up. If two files lack the same filename then in many cases I would not consider them to be duplicates. Seems that the quickest method would be to compare, in order:

  • filename
  • size
  • md5 checksum
  • byte contents

Do any methods do this? Look at duff, fdupes, rmlint, fslint, etc.

The following method was top-voted on commandlinefu.com: Find Duplicate Files (based on size first, then MD5 hash)

Can filename comparison be added as a first step, size as a second step?

find -not -empty -type f -printf "%s\n" | sort -rn | uniq -d | \
  xargs -I{} -n1 find -type f -size {}c -print0 | xargs -0 md5sum | \
  sort | uniq -w32 --all-repeated=separate
  • 3
    I've used duff, fdupes and rmlint, and strongly recommend readers to look at the third of these. It has an excellent option set (and documentation). With it, I was able to avoid a lot of the post-processing I needed to use with the other tools.
    – dubiousjim
    Commented Sep 2, 2015 at 6:32
  • 4
    In my practice filename is the least reliable factor to look at, and I've completely removed it from any efforts I make a de-duping. How many install.sh files can be found on an active system? I can't count the number of times I've saved a file and had name clash, with some on-the-fly renaming to save it. Flip side: no idea how many times I've downloaded something from different sources, on different days, only to find they are the same file with different names. (Which also kills the timestamp reliability.) 1: Size, 2: Digest, 3: Byte contents.
    – Chindraba
    Commented Jan 28, 2017 at 6:40
  • @GypsySpellweaver: (1) depends on personal use-case, wouldn't you agree? In my case, i have multiple restores from multiple backups, where files with same name and content exist in different restore-folders. (2) Your comment seems to assume comparing filename only. I was not suggesting to eliminate other checks.
    – johny why
    Commented Mar 8, 2017 at 21:50
  • 1
    Don't use rmlint. It emits a shellscript that doesn't check for errors. It deletes the duplicate then attempts to hardlink which can fail (e.g. due to too many links) and then continues to work on the next duplicate which suffers the same fate. It eats your files. The correct way would be to create it under a temporary name and then replace the dupe via rename.
    – the8472
    Commented Oct 7, 2020 at 10:01

I've used many of the hardlinking tools for Linux mentioned here. I too am stuck with ext4 fs, on Ubuntu, and have been using its cp -l and -s for hard/softlinking. But lately noticed the lightweight copy in the cp man page, which would imply to spare the redundant disk space until one side gets modified:

          control clone/CoW copies. See below

       When  --reflink[=always]  is specified, perform a lightweight copy, where the 
data blocks are copied only when modified.  If this is not possible the
       copy fails, or if --reflink=auto is specified, fall back to a standard copy.
  • I think I will update my cp alias to always include the --reflink=auto parameter now
    – Marcos
    Commented Mar 14, 2012 at 14:08
  • 1
    Does ext4 really support --reflink?
    – user12810
    Commented Jun 21, 2012 at 8:42
  • 7
    This is supported on btrfs and OCFS2. It is only possible on copy-on-write filesystems, which ext4 is not. btrfs is really shaping up. I love using it because of reflink and snapshots, makes you less scared to do mass operations on big trees of files.
    – clacke
    Commented Jul 3, 2012 at 18:57

Since I'm not a fan of Perl, here's a bash version:



find $DIR -type f -exec md5sum {} \; | sort > /tmp/sums-sorted.txt

for i in `cat /tmp/sums-sorted.txt`; do
 NEWSUM=`echo "$i" | sed 's/ .*//'`
 NEWFILE=`echo "$i" | sed 's/^[^ ]* *//'`
 if [ "$OLDSUM" == "$NEWSUM" ]; then
  echo ln -f "$OLDFILE" "$NEWFILE"

This finds all files with the same checksum (whether they're big, small, or already hardlinks), and hardlinks them together.

This can be greatly optimized for repeated runs with additional find flags (eg. size) and a file cache (so you don't have to redo the checksums each time). If anyone's interested in the smarter, longer version, I can post it.

NOTE: As has been mentioned before, hardlinks work as long as the files never need modification, or to be moved across filesystems.

  • How can I change your script, so that instead of hardlinking it, it will just delete the duplicate files and will add an entry to a CSV file the deleted file -> Lined File. . ???
    – MR.GEWA
    Commented Jan 12, 2013 at 12:17
  • Sure. The hard link line: echo ln -f "$OLDFILE" "$NEWFILE" Just replaces the duplicate file with a hard link, so you could change it rm the $NEWFILE instead.
    – seren
    Commented Jan 13, 2013 at 4:15
  • and how on next line, write in some text file somehow $OLDFILE-> NEWFILE ???
    – MR.GEWA
    Commented Jan 13, 2013 at 13:12
  • Ahh, right. Yes, add a line after the rm such as: echo "$NEWFILE" >> /var/log/deleted_duplicate_files.log
    – seren
    Commented Jan 14, 2013 at 19:28
  • 2
    Don't friggin reinvent the wheel. There are more mature solutions available, like rdfind, that works at native speeds and just requires brew install rdfind or apt-get install rdfind to get installed.
    – oligofren
    Commented Jan 3, 2015 at 13:46

I made a Perl script that does something similar to what you're talking about:


Basically, it just traverses a directory, calculating the SHA1sum of the files in it, hashing it and linking matches together. It's come in handy on many, many occasions.

  • 2
    I hope to get around to trying this soon... why not upload it on CPAN... App::relink or something Commented Feb 7, 2011 at 11:12
  • 2
    @xenoterracide: because of all the similar and more mature solutions that already exist. see the other answers, especially rdfind.
    – oligofren
    Commented Jan 3, 2015 at 13:36
  • 1
    @oligofren I don't doubt better solutions exist. TMTOWTDI I guess. Commented Jan 5, 2015 at 15:49

If you want to replace duplicates by Hard Links on mac or any UNIX based system, you can try SmartDupe http://sourceforge.net/projects/smartdupe/ am developing it

  • 6
    Can you expand on how “smart” it is? Commented Nov 4, 2012 at 13:25
  • 1
    How can I compare files of two different directories?
    – Burcardo
    Commented May 31, 2016 at 8:26

The applicatios FSLint (http://www.pixelbeat.org/fslint/) can find all equal files in any folder (by content) and create hardlinks. Give it a try!

Jorge Sampaio

  • 1
    It hangs scanning 1TB almost full ext3 harddisk, brings the entire system to a crawl. Aborted after 14 hours of "searching" Commented Sep 12, 2016 at 11:09

If you'll do hardlinks, pay attention on rights on that file. Notice, owner, group, mode, extended attributes, time and ACL (if you use this) is stored in INODE. Only file names are different because this is stored in directory structure, and other points to INODE properties. This cause, all file names linked to the same inode, have the same access rights. You should prevent modification that file, because any user can damage file to other. It is simple. It is enough, any user put other file in the same name. Inode number is then saved, and original file content is destroyed (replaced) for all hardlinked names.

Better way is deduplication on filesystem layer. You can use BTRFS (very popular last time), OCFS or like this. Look at the page: https://en.wikipedia.org/wiki/Comparison_of_file_systems , specialy at the table Features and column data deduplication. You can click it and sort :)

Specially look at ZFS filesystem. This is available as FUSE, but in this way it's very slow. If you want native support, look at the page http://zfsonlinux.org/ . Then you must patch kernel, and next install zfs tools for managament. I don't understand, why linux doesn't support as drivers, it is way for many other operating systems / kernels.

File systems supports deduplication by 2 ways, deduplicate files, or blocks. ZFS supports block. This means, the same contents that repeats in the same file can be deduplicated. Other way is time when data are deduplicated, this can be online (zfs) or offline (btrfs).

Notice, deduplication consumes RAM. This is, why writing files to ZFS volume mounted with FUSE, cause dramatically slow performance. This is described in documentation. But you can online set on/off deduplication on volume. If you see any data should be deduplicated, you simply set deduplication on, rewrite some file to any temporary and finally replace. after this you can off deduplication and restore full performance. Of course, you can add to storage any cache disks. This can be very fast rotate disks or SSD disks. Of course this can be very small disks. In real work this is replacement for RAM :)

Under linux you should take care for ZFS because not all work as it should, specialy when you manage filesystem, make snapshot etc. but if you do configuration and don't change it, all works properly. Other way, you should change linux to opensolaris, it natively supports ZFS :) What is very nice with ZFS is, this works both as filesystem, and volumen manager similar to LVM. You do not need it when you use ZFS. See documentation if you want know more.

Notice difference between ZFS and BTRFS. ZFS is older and more mature, unfortunately only under Solaris and OpenSolaris (unfortunately strangled by oracle). BTRFS is younger, but last time very good supported. I recommend fresh kernel. ZFS has online deduplication, that cause slow down writes, because all is calculated online. BTRFS support off-line dedupliaction. Then this saves performance, but when host has nothing to do, you run periodically tool for make deduplication. And BTRFS is natively created under linux. Maybe this is better FS for You :)

  • 1
    I do like the offline (or batch) deduplication approach btrfs has. Excellent discussion of the options (including the cp --reflink option) here: btrfs.wiki.kernel.org/index.php/Deduplication Commented Feb 5, 2017 at 19:42
  • ZFS is not Solaris or OpenSolaris only. It's natively supported in FreeBSD. Also, ZFS on Linux is device driver based; ZFS on FUSE is a different thing. Commented Mar 29, 2018 at 19:07

Easiest way is to use special program dupeGuru

dupeGuru Preferences Screenshot

as documentation says

Deletion Options

These options affect how duplicate deletion takes place. Most of the time, you don’t need to enable any of them.

Link deleted files:

The deleted files are replaced by a link to the reference file. You have a choice of replacing it either with a symlink or a hardlink. ... a symlink is a shortcut to the file’s path. If the original file is deleted or moved, the link is broken. A hardlink is a link to the file itself. That link is as good as a “real” file. Only when all hardlinks to a file are deleted is the file itself deleted.

On OSX and Linux, this feature is supported fully, but under Windows, it’s a bit complicated. Windows XP doesn’t support it, but Vista and up support it. However, for the feature to work, dupeGuru has to run with administrative privileges.


There is a new file level deduplication tool: https://gitlab.com/lasthere/dedup

On "BTRFS" and "XFS" it can can directly "reflink" identical files. While for non reflink supporting file systems, one first needs to produce a shell script via option "--print-only sh" and modify it such that it replaces the found identical files by a hardlink (cp --link).

It is especially useful if run regularly and file modifications (new dedupe candidates) are located separately, so that only that directory needs a deeper check, while all other locations are only scanned for finding deduplication sources.

(I'm author of dedup)


Hard links might not be the best idea; if one user changes the file, it affects both. However, deleting a hard link doesn't delete both files. Plus, I am not entirely sure if Hard Links take up the same amount of space (on the hard disk, not the OS) as multiple copies of the same file; according to Windows (with the Link Shell Extension), they do. Granted, that's Windows, not Unix...

My solution would be to create a "common" file in a hidden folder, and replace the actual duplicates with symbolic links... then, the symbolic links would be embedded with metadata or alternate file streams that only records however the two "files" are different from each other, like if one person wants to change the filename or add custom album art or something else like that; it might even be useful outside of database applications, like having multiple versions of the same game or software installed and testing them independently with even the smallest differences.

  • I'm afraid symlinks don't provide the functionality you write about. And "streams" are pure fantasy in this case. Commented Jun 21, 2022 at 20:30

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .